Moth Detection from Pheromone Trap Images Using Deep Learning Object Detectors

Abstract

Diverse pheromones and pheromone-based traps, as well as images acquired from insects captured by pheromone-based traps, have been studied and developed to monitor the presence and abundance of pests and to protect plants. The purpose of this study is to construct models that detect three species of pest moths in pheromone trap images using deep learning object detection methods and compare their speed and accuracy. Moth images in pheromone traps were collected for training and evaluation of deep learning detectors. Collected images were then subjected to a labeling process that defines the ground truths of target objects for their box locations and classes. Because there were a few negative objects in the dataset, non-target insects were labeled as unknown class and images of non-target insects were added to the dataset. Moreover, data augmentation methods were applied to the training process, and parameters of detectors that were pre-trained with the COCO dataset were used as initial parameter values. Seven detectors—Faster R-CNN ResNet 101, Faster R-CNN ResNet 50, Faster R-CNN Inception v.2, R-FCN ResNet 101, Retinanet ResNet 50, Retinanet Mobile v.2, and SSD Inception v.2 were trained and evaluated. Faster R-CNN ResNet 101 detector exhibited the highest accuracy (mAP as 90.25), and seven different detector types showed different accuracy and speed. Furthermore, when unexpected insects were included in the collected images, a four-class detector with an unknown class (non-target insect) showed lower detection error than a three-class detector.

Automatic Pest Counting from Pheromone Trap Images Using Deep Learning Object Detectors for Matsucoccus thunbergianae Monitoring

“…Recently, there have been many studies related to pest detection based on deep learning [14][15][16][17][18]. The studies are different in factors such as target insect type, size, imaging conditions, number of classes, and evaluation metrics; therefore, it is difficult to compare quantitative accuracy between studies.…”

“…In particular, for small-sized (1.5-2 mm) insects such as M. thunbergianae, there is a substantial difference in counting results because of identification skill and fatigue. To avoid labor-intensive manual counting, image-based automated monitoring methods have been used in many studies [9][10][11][12][13][14][15][16][17][18]. The image-based method captures images of traps and recognizes and counts pests on traps using image processing.…”

“…Nieuwenhuizen et al [16] used Faster R-CNN to detect trapped insects, and Chulu et al [17] identified and classified fall army worm moths based on CNN. In our previous study [18], we applied several deep learning object detection models, such as Faster R-CNN, R-FCN, Retinanet, and SSD to detect three types of moths in trap images and compared the speed and performance of each model.…”

Development of an Automated Multidirectional Pest Sampling Detection System Using Motorized Sticky Traps

“…Some studies [3,4] have chosen insect specimens that have been well preserved in an ideal laboratory environment to avoid problems, and images can be captured at high resolution; however, given that less environmental issues are considered in this method, they are limited in specific applications. A more realistic approach is to use traps, which attract insects by several ways, based on light [5], color [6,7], and pheromones [8,9].…”

“…Sticky traps usually consist of a piece of paper containing a sticky substance to trap insects and are placed in fields or greenhouses [13,14]. Sticky traps are then photographed for automatic counting [7,9,13]. This strategy offers a smooth and neutral surface, making insect counting easier.…”